Why 'open' AI systems are actually closed, and why this matters

Area of Science:

• Computer Science
• AI Ethics
• Science and Technology Studies

Background:

• The definition and implications of 'open' artificial intelligence (AI) are debated.
• Current discourse often misapplies open-source software concepts to AI.
• Industry actors leverage 'open' AI claims to influence policy regarding innovation, democracy, and safety.

Purpose of the Study:

• To critically examine the basis and implications of 'open' AI claims.
• To provide a material analysis of what constitutes openness in AI systems.
• To clarify the relationship between AI openness and power concentration.

Main Methods:

• Analysis of AI development and deployment practices.
• Examination of AI models, data, labor, frameworks, and computational resources.
• Comparative analysis with open-source software models.

Main Results:

• 'Open' AI offers affordances like transparency, reusability, and extensibility.
• Maximal openness allows for some oversight and experimentation.
• Openness alone does not disrupt the concentration of power in the AI sector.

Conclusions:

• Rhetoric around 'open' AI can be used to consolidate rather than reduce power concentration.
• The co-option of open-source software by large companies offers a parallel to current AI trends.
• Policy definitions of 'openness' in AI are critical and require careful scrutiny.